Foxtech Provides Industrial Drone Solutions & UAV Payload Systems.
Now Launched: Agility A2 + ExoArm-7 VLA Suite for Embodied AI Research
Building capable robots starts with better human demonstrations.
The new Agility A2 + ExoArm-7 VLA Suite is a wearable bimanual teleoperation platform designed to help researchers, developers, and robotics teams move faster from human demonstration to robot learning and real-world task validation.
By combining the Agility A2 dual-arm robot with the ExoArm-7 wearable exoskeleton system, this solution enables natural human motion capture, synchronized multimodal data collection, model training, and real-time robot inference within one integrated workflow.
For embodied AI research, imitation learning, and Vision-Language-Action development, it provides a more direct path from “showing” a robot how to perform a task to building an autonomous policy that can execute it.
What problem does the Agility A2 + ExoArm-7 VLA Suite solve?
Collecting high-quality robot-learning data is often slow and difficult.
Traditional teleoperation setups may require separate controllers, external tracking devices, manually synchronized cameras, and custom software pipelines. This increases setup time and makes it harder to capture consistent demonstrations for AI training.
The Agility A2 + ExoArm-7 VLA Suite brings these elements together in one preconfigured platform.
Operators wear the ExoArm-7 exoskeleton and control the robot through natural arm movements. The system captures synchronized human motion, robot state information, RGB-D visual data, and point clouds, creating a richer dataset for imitation learning and VLA-based robot training.
Instead of building a data-collection workflow from scratch, teams can focus on the task itself: grasping, sorting, insertion, assembly, transport, and other real manipulation scenarios.
How does wearable teleoperation improve human demonstration capture?
The ExoArm-7 wearable exoskeleton is built for more natural and intuitive bimanual control.
With 14 degrees of freedom and motion-capture latency below 20 ms, the wearable system maps the operator’s arm movements to the Agility A2 dual-arm robot in real time. This makes teleoperation feel closer to demonstrating a task directly rather than controlling a robot through a conventional remote interface.
For researchers, this means human demonstrations can be collected with greater continuity and precision.
For robot-learning workflows, it means the robot can capture not only visual information, but also the motion patterns behind each action: how an object is approached, how the arms coordinate, how a grasp is adjusted, and how a task is completed.
These high-fidelity demonstrations can then support imitation learning, policy training, reinforcement learning experiments, and VLA dataset generation.
What can the dual-arm robot do?
At the core of the suite is the Agility A2 dual-arm robot, featuring two 7-DOF robotic arms for human-like bimanual manipulation.
Each arm supports a 5 kg rated payload and up to a 12 kg peak payload, allowing the platform to handle a broad range of research and validation tasks. From picking and placing to assembly, insertion, handling, and coordinated two-arm operations, the system is designed for practical manipulation research rather than single-task demonstrations only.
The dual-arm design is especially valuable for tasks that require coordination between both hands, such as stabilizing an object with one arm while operating it with the other.
This makes the platform well suited for studying how robots can learn complex manipulation skills from human demonstrations.
How does the system capture visual and spatial data?
Robot learning requires more than movement data alone.
The Agility A2 + ExoArm-7 VLA Suite includes an Intel D435 head camera and dual Intel D405 wrist cameras integrated with the vision gripper assemblies. Together, these cameras provide synchronized RGB-D data and point-cloud information from both overview and close-range manipulation perspectives.
The head camera captures the broader workspace, while the wrist cameras observe critical interactions near the grippers.
This multi-camera approach helps create richer datasets for visual perception, object understanding, grasp planning, and action learning. It also gives researchers more flexibility when studying how robot policies respond to changing object positions, workspace layouts, and manipulation conditions.
Can it support a complete VLA workflow?
Yes. The platform is designed as an end-to-end embodied AI workflow rather than only a teleoperation device.
Powered by the open-source LeRobot framework, the suite supports human demonstration capture, synchronized data acquisition, dataset generation, training workflows, model inference, and autonomous task execution.
The system comes with an OpenarmX-LeRobotVLA preconfigured software environment, helping teams begin development with less integration work.
ACT inference is supported by default. With higher-performance controller configurations, the platform also supports advanced VLA models including SmoLVLA, Pi0, Pi0.5, and XVLA.
This gives researchers a practical environment for testing different learning approaches, comparing policies, and validating whether an AI model can transfer from collected demonstrations to real robot execution.
What development tools and ecosystems are supported?
The Agility A2 + ExoArm-7 VLA Suite is built for teams that need flexibility across research, simulation, and deployment.
It supports a broad robotics development ecosystem, including:
- ROS2 and ROS2-control
- NVIDIA Isaac Sim
- MuJoCo
- MoveIt motion planning
- URDF robot modeling
- Python and C++ development
- Inverse kinematics and trajectory optimization
With gravity compensation, CANFD communication, high-speed USB expansion, and a dedicated control workstation, the system provides a stable foundation for synchronized sensors, real-time control, and repeatable motion experiments.
The standard workstation is equipped with an AMD Ryzen 9 Pro processor, 16 GB RAM, and a 1 TB SSD, with higher configurations available for more demanding VLA workloads.
Who is this platform built for?
The Agility A2 + ExoArm-7 VLA Suite is ideal for:
- Universities and robotics education programs
- Embodied AI and robot-learning laboratories
- AI researchers building imitation-learning or VLA datasets
- Industrial R&D teams validating manipulation workflows
- Robotics startups developing rapid proof-of-concept demonstrations
- Teams exploring simulation-to-real robot learning
Whether the goal is collecting demonstration data, teaching dual-arm manipulation, evaluating new policies, or rapidly validating a robot-learning concept, the system offers an integrated path from human motion to autonomous execution.
Why choose the Agility A2 + ExoArm-7 VLA Suite?
Embodied AI development requires more than a capable robot arm. It requires a reliable way to collect meaningful demonstrations, connect visual perception with action, train learning models, and test the results in the real world.
The Agility A2 + ExoArm-7 VLA Suite brings these critical steps together in one wearable dual-arm teleoperation platform.
From natural human motion capture to synchronized RGB-D vision, from VLA dataset generation to autonomous robot inference, it is designed to help robotics teams reduce integration time and accelerate experimentation.
Turn human demonstration into robot capability — with the Agility A2 + ExoArm-7 VLA Suite.
